jason

R中如何做glm

#data importation 

model <- glm(y ~ ., data = wave, family=binomial)

model.default <- glm(y ~ 1, data = wave, family=binomial) 

model.forward <- 

stepAIC(model.default,

scope=list(lower=as.formula(model.default),upper=as.formula(model)),

direction="forward",

k=log(nrow(wave))

)

说明这里的变量选择用了stepAIC, optimizes the Akaike (AIC) criterion.

 According the forward search, we try all the 

regression with 1 predictor. We perform "p" logistic regression learning process. Then, we select the best one. We try to add a second variable. So, we perform "p-1" regressions. The algorithm is  quadratic O(p²). The calculation time is heavily impacted.